| Source DB | nl |
|---|
| Institution | UGent |
|---|
| Code | fb9e52e1-2769-4915-8a67-715a23554377 |
|---|
| Unit | 3d80f44d-a6bd-4650-a26f-a7d882b99d2b
|
|---|
| Begin | 10/1/2018 |
|---|
| End | 9/2/2020 |
|---|
| title fr |
|
|---|
| title nl | Snelheid en accuraatheid combineren in computationele chemie: Machinaal leren voor korte-dracht interacties aangevuld met fysische modellen op lange-afstand
|
|---|
| title en | Combining speed and accuracy in computational chemistry: Machine learning for short-range interactions augmented by physical models at long-range
|
|---|
| Description fr |
|
|---|
| Description nl | Statistical-learning approaches are emerging as powerful alternatives to expensive computationalmethods for solving the Schrödinger equation to determine molecular properties. Despite therecent success of methods like neural networks, these models are only suitable for interpolationand fail to scale to larger systems. That is, when a model is trained on small-to-medium-sizemolecules, it can only be applied to systems of similar size. Modeling long-range intermolecularinteractions with machine learning (ML) requires sampling the vast diversity of chemicalenvironments occurring on an extended length scale. This leads to a combinatorial explosion in therequired amount of training data.To circumvent these obstacles, I propose to incorporate our physical knowledge of long-rangeinteractions into the modeling process; this is philosophically different from the commonly usedU+201Cblack box ML modelingU+201D. A detailed analysis of the proposed model reveals that it can achieve theaccuracy of high-level quantum chemistry at the cost of molecular mechanics. This not only allowsone to compute interaction energies of large molecules (e.g. drug-target binding) and run longtimemolecular dynamics simulations of (macro)molecules, but also enables accurate and efficientcomputational screening of large databases to select the most promising molecules for follow-upexperiments. Besides its transformative utility, this pioneering strategy is extendable to manyother problems in chemistry.
|
|---|
| Description en | Statistical-learning approaches are emerging as powerful alternatives to expensive computationalmethods for solving the Schrödinger equation to determine molecular properties. Despite therecent success of methods like neural networks, these models are only suitable for interpolationand fail to scale to larger systems. That is, when a model is trained on small-to-medium-sizemolecules, it can only be applied to systems of similar size. Modeling long-range intermolecularinteractions with machine learning (ML) requires sampling the vast diversity of chemicalenvironments occurring on an extended length scale. This leads to a combinatorial explosion in therequired amount of training data.To circumvent these obstacles, I propose to incorporate our physical knowledge of long-rangeinteractions into the modeling process; this is philosophically different from the commonly usedU+201Cblack box ML modelingU+201D. A detailed analysis of the proposed model reveals that it can achieve theaccuracy of high-level quantum chemistry at the cost of molecular mechanics. This not only allowsone to compute interaction energies of large molecules (e.g. drug-target binding) and run longtimemolecular dynamics simulations of (macro)molecules, but also enables accurate and efficientcomputational screening of large databases to select the most promising molecules for follow-upexperiments. Besides its transformative utility, this pioneering strategy is extendable to manyother problems in chemistry.
|
|---|
| Qualifiers | - Machinaal leren - Machine learning - |
|---|
| Personal | Heidar Zadeh Farnaz, Verstraelen Toon |
|---|
| Collaborations | |
|---|
